-
Notifications
You must be signed in to change notification settings - Fork 0
Home
“As a foundation of the Nova Scotia Health Authority, we are working to place the dignity and respect of patients, families and communities at the heart of every decision. We seek to build trust-based relationships to achieve more genuine partnerships with those we serve.”
Founding Principles: Medex
Initial Hypothesis: I am going to combine the latest therapies and counselling strategies, with the latest (and one of the largest) medical language models. It will be called Med-Exec, your Medical Executive Assistant for medical relationships. The main feature is you can add all your personal medical information (supported in a variety of formats, which can be diagnostic), then ask the assistant questions about it. With reference to the internet’s vast information, it will chat in a responsive but informative way. Users can select information from that text which they may want to share for informational purposes with family or friends. That way, they can add peers individually like WhatsApp, then those people can talk to your assistant directly and only about the information you want them to know about. You can also add your doctors, which will allow them to review other Doctors’ notes.
An important distinction is that the app is not designed for communicating. It’s designed so that your assistant can put together puzzle pieces that may be emotionally challenging, and/ or physically difficult to interpret and then relay in comprehensive way.
“Medical language is complex, and when information is relayed- the important complexities of medical conversations get lost in the over-simplification of trying to relay what was just explained by a medical professional.” - Matthew Simon
That way, everyone in your circle can learn in their own way, at their own level, and come together later to discuss outside of the app, and once your assistant has helped get everyone on the same page. If your circle feels they don’t know how to discuss your situation, the flip side is the application is also trained on the latest counselling and therapy techniques. So you can just ask it what to ask/ say/ talk about, etc. making it easier for everyone to communicate in a unified language while also having a safe place they can ask their questions to the assistant, rather than potentially mis-speaking in a potentially vulnerable situation.
The goal of Med-Exec is to provide a safe place to hold your medical history, learn about it, and share it in ways that feel less vulnerable, and give others the opportunity to learn in their own way.
“I believe one of the hardest parts about being sick is simply having to explain it to family. - Matthew Simon”
Medex 'black-box' will function via the 5-steps below:
-
User Query Understanding: Analyze the user query using NLP techniques. The goal is to determine the user's level of medical understanding, identify key medical terms, and understand the sentiment of the query. Machine learning models can be trained to identify different levels of complexity in language and classify the user's query into one of the five categories you've specified.
-
User Healthcare Cycle Identification: Similar NLP models can be used to determine the user's stage in the healthcare cycle. Keywords or phrases in the user's query can hint at whether they are pre-care, currently under care, or post-care.
-
Medical Language Translation: With the user's comprehension level and healthcare stage determined, the system could then rephrase or translate the query into more medically complex language. This step would require a vast knowledge base of medical terms and concepts, potentially aided by medical ontology resources such as SNOMED CT, UMLS, or MeSH.
-
Question Expansion: The medically translated query can then be used to generate related questions. Techniques such as query expansion in information retrieval can be useful here. This would involve using synonymy and semantic relationships between medical terms to create multiple related queries.
-
Embedding Database Search: Once you have a set of expanded, medically translated queries, these can be compared with the documents in your database using a technique like vector space modeling. Document embeddings (representations of documents as multi-dimensional vectors) can be compared with the query vector to find the most relevant results.